Background
The 2009 Machel report estimates that just over one billion children and adolescents live in countries and territories affected by armed conflict [
1]. In 2011 alone, 37 armed conflicts were recorded globally, the majority in Africa (n = 15, 41%), Asia (n = 13, 35%), and the Middle East (n = 6, 16%) [
2]. Epidemiological studies have shown that armed conflicts are associated with a wide range of child mental health outcomes. These may range from resilience, that is, maintained mental health in the face of adversity, to increased psychological distress and heightened prevalence of mental disorders including (symptoms of) post-traumatic stress disorder (PTSD), depression, and anxiety disorders [
3].
To address the mental health burden in humanitarian settings, mental health and psychosocial support interventions are increasingly popular and consensus-based guidelines for such interventions have been developed [
4,
5]. These guidelines recommend implementing multi-layered packages of services, including preventive and treatment interventions, to take into account the diversity of mental health and psychosocial needs in humanitarian settings. The current study concerns a school-based mental health intervention implemented within a multi-layered package of services [
6,
7]. Within this package, the school-based intervention was aimed both at reducing psychological symptoms (treatment aim), as well as improving strengths and functioning in children with heightened symptomatology (preventive aim).
Despite consensus on best practices, little rigorous evidence is available on the effectiveness of child mental health interventions in humanitarian settings [
8,
9]. A recent meta-analysis of interventions with children affected by armed conflict in low- and middle-income countries, including six randomized controlled trials with no-intervention comparison groups, showed high heterogeneity of intervention effects across studies [
10]. This high heterogeneity may be due to the diversity of interventions included in the meta-analysis (that is, specialized psychotherapeutic interventions and preventive interventions), but may also be associated with individual and contextual factors that influence intervention effects. To improve knowledge on what works for whom and under what circumstances, a crucial research direction is the identification of mediators and moderators of interventions. Mediators are variables that identify why and how interventions have effects, whereas moderators are variables that identify on whom and under what circumstances interventions have different effects [
11]. Identification of mediators and moderators may assist in adapting interventions to make them more effective, or identifying the populations and contexts for which interventions are most beneficial.
This study was implemented with conflict-affected children in Burundi. Burundi is a landlocked country in the Great Lakes region of eastern Africa, with a population of 8.5 million. It is one of the poorest countries of the world, ranking 185 out of 187 countries on the Human Development Index [
12]. The country has experienced cyclical ethnic violence between Hutu and Tutsi ethnic groups since 1962. The most recent violence occurred in 1993, when the killing of the first elected Hutu president sparked a civil war that killed 300,000 people and displaced 1.2 million people. Although a peace agreement was signed by most warring parties in 2000, political instability and violence continued up to the time of the study [
13].
We aimed to address three research questions. First, our treatment aim: What is the effectiveness of a school-based intervention to reduce psychological symptoms (primary outcomes: PTSD, depressive, and anxiety symptoms)? Second, our preventive aim: What is the efficacy of a school-based intervention to improve hope and improve functioning (secondary outcomes)? Finally, we wanted to address the question: What are the mediators and moderators of intervention outcomes? We hypothesized that the intervention would be associated with greater reductions in symptomatology and function impairment, as well as greater improvements of a sense of hope. Our hypotheses of mediators and moderators were based on the theoretical notion of ‘ecological resilience,’ that is we expected that intervention effects would be determined by protective and risk factors at various levels of children’s social ecology (individual, family, peer, community) [
14,
15]. With regard to mediators, we were interested in coping and social support. We hypothesized that the intervention would be associated with larger improvements in coping and social support among children in the intervention condition, and that these improvements in turn would be associated with improvements on PTSD, depressive, and anxiety symptoms, and hope and function impairment. A systematic review on resilience and mental health in children affected by armed conflict found various studies supporting a relationship between coping and social support and lower levels of psychological symptoms, although these relations were often symptom-specific and varied by phase of conflict [
16].
With regard to moderators, we hypothesized that intervention effects would vary by gender and age. Previous evaluation studies of psychosocial interventions in diverse settings have found differing effects by age and gender [
17‐
21]. In addition, we were interested in the moderating roles of family-level variables, including household size, family connectedness, displacement status, and family composition. A longitudinal study with Afghan children found that quality of family life was an important predictor of psychological symptoms over time [
22], as did a cross-sectional study in Lebanon [
23]. An evaluation of a psychosocial intervention in conflict-affected areas in Indonesia found that household size influenced size of intervention effects [
24]. Finally, we were interested in the potential moderating role of community-level variables, that is, social capital. In a longitudinal study with former child soldiers in Sierra Leone, a different but related variable (community acceptance) was associated with higher levels of prosocial behavior and lower levels of internalizing and externalizing symptoms over time [
25]. Although the impact of armed conflict on supportive community relations has been a frequent theme in the literature on children and armed conflict [
26], we are not aware of studies that have examined the role of social capital as a moderator of intervention efficacy. We hypothesized that children who perceived low levels of social capital would report stronger benefits from a psychosocial intervention, given its focus on improving supportive relations between peers.
Discussion
This study was aimed at identifying intervention outcomes, and mediators and moderators of a school-based mental health intervention for children in war-affected Burundi. The intervention was aimed both at the reduction of PTSD, depressive, and anxiety symptoms (treatment aim, primary outcomes), as well as the improvement of hope and functioning (preventive aim, secondary outcomes). We did not find any main effects on the primary and secondary outcome measures, that is, for either the treatment or preventive aims. Therefore, mediation analyses were not performed. However, we did find eight differences in longitudinal trajectories between study conditions in interaction with a number of moderators. These moderation effects are challenging to interpret clinically, as commonly applied effect size calculations do not take clustered variance into account and may obscure subgroup findings. However, moderation effects generally applied to larger groups of children in our sample (for example, children living with both parents, 67.2% of the sample; children with larger households, 54.3% had seven members or more; children with lower trauma exposure, 21.6% reported two or fewer events; children living in their original village or bought land, 54.8%).
Before discussing these findings in more detail, we highlight limitations of the study. First, we found statistically significant differences between the study conditions at baseline, which may have been the result of randomization by province before randomization of schools. It is unknown how these differences may have impacted findings on intervention effects, given that children in the intervention condition were doing better in some aspects (that is, lower depressive symptoms, higher social capital) but worse in other aspects (that is, higher trauma exposure, fewer coping strategies and coping satisfaction, lower social support) at baseline. Differences may also have been associated with the fact that relatively few schools were randomized, which the analyses controlled for. In addition, differences were found mainly for variables for which we did not find significant intervention effects, with the exception of trauma exposure and depressive symptoms (three of eight differences identified between study conditions). Second, drop-out may have affected study findings. As with the baseline differences between study conditions, it is difficult to say how these findings may have impacted study findings, because study completers had lower levels of PTSD and higher levels of social capital, but lower levels of social support. Third, we had to exclude our measure of anxiety due to low IR of the measure. Although our other measures had good IR and TRR, unmeasured factors that contributed to low IR of the anxiety measure may have impacted our other instruments in ways we did not measure. Fourth, our research assessors were working independently from the implementation team, but were not blinded to study condition because they had to visit schools to interview children. We emphasized in training of assessors that an objective evaluation was crucial. However, the lack of blinding assessors may have biased findings in favor of the intervention. Strengths of the study include the detailed translation and mixed methods procedures to prepare instruments, the inclusion of a broader range of measures to assess different aspects of mental health (symptoms, hope, functioning), the inclusion of a follow-up assessment three months after the intervention, and the examination of moderators.
Our current findings from Burundi add to a number of recent studies that have rigorously evaluated this school-based intervention with conflict-affected populations in Indonesia, Nepal, the occupied Palestinian territories, and Sri Lanka. Collectively, these studies provide emerging evidence-based answers to important questions on the practical benefit of this and similar interventions. First, given inconsistent results on primary outcome measures across settings, it now seems that this school-based intervention should not be recommended as a treatment for PTSD, depressive, and anxiety symptoms. Although the school-based intervention was associated with reductions in these symptoms in some settings (such as girls in Indonesia, boys and children with less ongoing trauma exposure in Sri Lanka, and children with both parents in Burundi), it was not associated with improvements in these symptoms in Nepal and the occupied Palestinian territories, and was associated with unfavorable effects on PTSD symptoms in Sri Lankan girls and older children in the occupied Palestinian territories. Rather, recent World Health Organization guidelines for non-specialized health settings in low- and middle-income countries recommend cognitive behavioral treatments with a trauma focus and eye movement desensitization reprocessing as treatments for PTSD in children and adolescents [
45]. Although this school-based intervention incorporated cognitive behavioral elements (for example, working on coping skills, psycho-education, some discussion of trauma-related material through drawings), it did not comprise consistent trauma exposure, memory, or cognitive processing when compared to cognitive behavioral treatments with a trauma focus or to eye movement desensitization reprocessing. On the basis of findings in Indonesia and Sri Lanka, we previously argued that this intervention may nevertheless have a place in a spectrum of treatment options: despite smaller main effects, it can reach more children with fewer resources. Given the inconsistent results across settings, however, it may be better to start with World Health Organization recommendations for treatment of PTSD in children and adolescents. A remaining key research question here is how existing evidence-based interventions, often tested in more highly resourced research settings, can be effectively disseminated and implemented in real-world health-care settings [
46‐
48].
When considering the benefit of the school-based intervention as a preventive intervention, that is, in strengthening resilience processes in conflict-affected children, the intervention seems to have more consistent results across settings. Intervention effects were found for hope, positive coping, social support, and function impairment in Indonesia; hope and prosocial behavior in Nepal; hope in Sri Lanka; hope and a range of other strengths in the occupied Palestinian territories; and hope and function impairment in this study. However, in this study in Burundi, displaced children in the intervention condition had worse trajectories on hope and function impairment. Further adaptation of this school-based intervention may focus on removing the trauma-focused elements (see [
49]) and implementing it only as a preventive tool, as well as concentrating on more active involvement of families and communities. An important future direction would then be to examine whether changes in strengths in the shorter term translate to improvements in psychological symptoms and overall wellbeing and development in the longer term.
An important question concerns the differential intervention effects (both treatment and preventive) by gender, age, and a variety of contextual factors. We feel results across studies may best be explained from the theoretical perspective of ecological resilience. This theoretical framework aims to explain children’s mental health by examining which resources (strengths) are available in children and their social contexts, at family, peer, and community levels. From this perspective, the complex differential effects of the intervention may be clarified by the extent to which the resources in children’s environment may interact with intervention activities. The most positive intervention effects were observed in Central Sulawesi, Indonesia, which may be explained by the fact that children there were still living in generally supportive families and communities - although tension remained between the communities after the conflict [
15,
33]. In Indonesia, the children that benefited most from the intervention were the children who were socially more isolated [
24]. It appears that in settings that are more volatile (for example, Burundi, Sri Lanka), the resilience of children that live in particularly stressful conditions may actually be undermined by the intervention (for example, girls and children exposed to higher levels of ongoing stressors in Sri Lanka, male adolescents in the occupied Palestinian territories, displaced children in Burundi). In these settings, a preliminary intervention recommendation would be to have a clearer separation between intervention aims and to implement interventions in more homogenous groups, so that only the positive preventive effects may be achieved with children in more stable situations, and more intensive treatments may reach those who are more vulnerable. Detailed pre-intervention assessments seem crucial to identify who the particularly socially vulnerable children may be.
From a scientific point of view, the identified differential effects from our studies call for a more detailed look at how context interacts with intervention effects in conflict-affected settings, particularly for evaluations of preventive interventions. This would require an adaptation of the randomized controlled trial (for example, with a cohort multiple randomized controlled trial design) to encompass the multi-disciplinary examination of family-, peer-, and community-level variables before and during implementation of interventions, and the study of how such variables interact over the longer term with psychological symptoms. Following advances that have been made in the field of treatment of PTSD symptoms, such scientific developments would aid in strengthening efforts to prevent long-lasting psychological symptoms and promote mental health in children affected by armed conflict.
Furthermore, a promising research direction for preventive efforts could be to intervene in earlier developmental periods, when parenting skills and parental mental health can be enhanced before negative patterns set in [
50]. Given that development proceeds sequentially (that is, skills learned later in life build on skills learned earlier), researchers focusing on children growing up in adversity have emphasized the early childhood period as a particularly cost-effective period for intervention [
51,
52]. For example, in a randomized controlled trial with 87 displaced mother-child dyads (mean age of children, 5.5 years), Dybdahl found that a group intervention aimed at strengthening parental involvement, support, and education in mothers had promising benefits for their children [
53].
Author contributions
WT, IK, MJ, RM, JdJ designed the study. WT, AN, ES supervised data collection. MJ and PN supervised implementation of the intervention. WT, HS analyzed the data, with critical inputs from IK, MJ, JdJ. WT wrote a first draft of the manuscript. WT, IK, MJ, AN, PN, HS, ES, RM, JdJ revised the manuscript for important intellectual content. All authors read and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.